Improved Structural Rotor Fault Diagnosis using Multi-sensor Fuzzy Recurrence Plots and Classifier Fusion

Rotating machinery (RM) fault diagnosis based on artificial intelligence (AI) is an esteemed industrial IoT application and an inevitable constituent of the Industry 4.0 revolution. Notably, the existing research in this area exhibits the following limitations: i) the time-series characterization or system dynamics considerations of vibration input are not strictly followed in the diagnostic procedure, ii) fault-specific component usage is not encouraged in existing decision-making strategies, and iii) single classifier based decision-making commonly compromises in utilizing all features of input data. Structural rotor fault (SRF) is the most vital but least attended defect, which is the root cause of most RM issues. We develop a framework that uses two parallel decision-making strategies, i.e., the convolutional neural network (CNN) with fuzzy recurrence plot (FRP) as input and a long short-term memory (LSTM) network taking distinctive frequency components (DFC) stream as input. Moreover, a DFC-based ranking and image combining scheme is proposed to select the most informative sensor signals from the segmented raw vibration signals and generate a 3D-FRP, and 2D flattened FRP (F-FRP). These multi-sensor FRPs characterize the system dynamics and preserve the time-series properties of the vibration input. The DFC stream simultaneously catered to LSTM facilitates symptom parameters-based decision-making. The parallel decision scores are fused by fuzzy integral (FI) based fusion depending on the confidence of the sources. The model performs exceptionally well over the state-of-the-art methods and achieves around 99.0% accuracy with the Meggitt testbed dataset and the MaFaulDa dataset.